Process mining on databases: Unearthing historical data from redo logs

Process Mining techniques rely on the existence of event data. However, in many cases it is far from trivial to obtain such event data. Considerable efforts may need to be spent on making IT systems record historic data at all. But even if such records are available, it may not be possible to derive an event log for the case notion one is interested in, i.e., correlating events to form process instances may be challenging. This paper proposes an approach that exploits a commonly available and versatile source of data, i.e. database redo logs. Such logs record the writing operations performed in a general-purpose database for a range of objects, which constitute a collection of events. By using the relations between objects as specified in the associated data model, it is possible to turn such events into an event log for a wide range of case types. The resulting logs can be analyzed using existing process mining techniques.